Complexity-Based Measures Inform Effects of Tai Chi Training on Standing Postural Control:...

RESEARCH ARTICLE

Complexity-Based Measures InformEffects of Tai Chi Training on StandingPostural Control: Cross-Sectional andRandomized Trial StudiesPeter M. Wayne1*, Brian J. Gow1, Madalena D. Costa2, C.-K. Peng2,3,Lewis A. Lipsitz4,5, Jeffrey M. Hausdorff 6,7, Roger B. Davis8, Jacquelyn N. Walsh1,Matthew Lough5, Vera Novak4,9, Gloria Y. Yeh8, Andrew C. Ahn8,10,Eric A. Macklin11, Brad Manor4

1. Osher Center for Integrative Medicine, Brigham and Women’s Hospital, Harvard Medical School, Boston,Massachusetts, United States of America, 2. Division of Interdisciplinary Medicine and Biotechnology andMargret and H.A. Rey Institute for Nonlinear Dynamics in Medicine, Beth Israel Deaconess Medical Center,Harvard Medical School, Boston, Massachusetts, United States of America, 3. Center for DynamicalBiomarkers and Translational Medicine, National Central University, Chungli, Taiwan, 4. Division ofGerontology, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, Massachusetts, UnitedStates of America, 5. Institute for Aging Research, Hebrew SeniorLife, Roslindale, Massachusetts, UnitedStates of America, 6. Movement Disorders Unit, Department of Neurology, Tel Aviv Medical Center, Tel Aviv,Israel, 7. Department of Physical Therapy and Sagol School of Neuroscience, Tel Aviv University, Tel Aviv,Israel, 8. Division of General Medicine and Primary Care, Beth Israel Deaconess Medical Center, HarvardMedical School, Boston, Massachusetts, United States of America, 9. Department of Neurology, Beth IsraelDeaconess Medical Center, Harvard Medical School, Boston, Massachusetts, United States of America, 10.Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Boston, Massachusetts, UnitedStates of America, 11. Biostatistics Center, Massachusetts General Hospital, Harvard Medical School,Boston, Massachusetts, United States of America

*[email protected]

Abstract

Background: Diminished control of standing balance, traditionally indicated by

greater postural sway magnitude and speed, is associated with falls in older adults.

Tai Chi (TC) is a multisystem intervention that reduces fall risk, yet its impact on

sway measures vary considerably. We hypothesized that TC improves the

integrated function of multiple control systems influencing balance, quantifiable by

the multi-scale ‘‘complexity’’ of postural sway fluctuations.

Objectives: To evaluate both traditional and complexity-based measures of sway

to characterize the short- and potential long-term effects of TC training on postural

control and the relationships between sway measures and physical function in

healthy older adults.

Methods: A cross-sectional comparison of standing postural sway in healthy TC-

nal̈ve and TC-expert (24.5¡12 yrs experience) adults. TC-nal̈ve participants then

completed a 6-month, two-arm, wait-list randomized clinical trial of TC training.

OPEN ACCESS

Citation: Wayne PM, Gow BJ, Costa MD, Peng C-K, Lipsitz LA, et al. (2014) Complexity-BasedMeasures Inform Effects of Tai Chi Training onStanding Postural Control: Cross-Sectional andRandomized Trial Studies. PLoS ONE 9(12):e114731. doi:10.1371/journal.pone.0114731

Editor: Heiner K. Berthold, Bielefeld EvangelicalHospital, Germany

Received: May 21, 2014

Accepted: November 10, 2014

Published: December 10, 2014

Copyright: � 2014 Wayne et al. This is an open-access article distributed under the terms of theCreative Commons Attribution License, whichpermits unrestricted use, distribution, and repro-duction in any medium, provided the original authorand source are credited.

Data Availability: The authors confirm that all dataunderlying the findings are fully available withoutrestriction. Ethical restrictions prevent public shar-ing of data. Data are available upon request fromDr. Peter Wayne ([email protected]).

Funding: This publication was made possible bygrant number R21 AT005501-01A1 from theNational Center for Complementary and AlternativeMedicine (NCCAM: http://nccam.nih.gov/) at theNational Institutes of Health (NIH: http://www.nih.gov/), and from grant number UL1 RR025758supporting the Harvard Clinical and TranslationalScience Center, from the National Center for

http://www.nih.gov/).

Its contents are solely the responsibility ofthe authors and do not necessarily represent theofficial views of the NCCAM, NCRR, or the NIH. Dr.Lipsitz was supported by grant R37-AG25037 fromthe National Institute on Aging (NIA: http://www.nia.nih.gov/) and the Irving and Edyth S. Usen andFamily Chair in Geriatric Medicine at HebrewSeniorLife. Dr. Novak has received grants from theNIH–National Institute of Diabetes and Digestiveand Kidney Diseases (NIDDK: http://www.niddk.nih.gov/) (5R21-DK-084463-02) and the NationalInstitutes of Health (NIH)–National Institute onAging (NIA: http://www.nia.nih.gov/) (1R01-AG-0287601-A2) unrelated to this study. The work inthe SAFE lab was conducted with support fromHarvard Catalyst (http://catalyst.harvard.edu/) | TheHarvard Clinical and Translational Science Center(National Center for Research Resources and theNational Center for Advancing TranslationalSciences, National Institutes of Health Award8UL1TR000170-05 and financial contributions fromHarvard University and its affiliated academic

PLOS ONE | DOI:10.1371/journal.pone.0114731 December 10, 2014 1 / 28

Research Resources (NCRR:about/almanac/organization/NCRR.htm

http://crossmark.crossref.org/dialog/?doi=10.1371/journal.pone.0114731&domain=pdf

http://creativecommons.org/licenses/by/4.0/

http://www.nih.gov/about/almanac/organization/NCRR.htm

http://www.nih.gov/about/almanac/organization/NCRR.htm

Postural sway was assessed before and after the training during standing on a

force-plate with eyes-open (EO) and eyes-closed (EC). Anterior-posterior (AP) and

medio-lateral (ML) sway speed, magnitude, and complexity (quantified by

multiscale entropy) were calculated. Single-legged standing time and Timed-Up–

and-Go tests characterized physical function.

Results: At baseline, compared to TC-nal̈ve adults (n560, age 64.5¡7.5 yrs), TC-

experts (n527, age 62.8¡7.5 yrs) exhibited greater complexity of sway in the AP

EC (P50.023), ML EO (P,0.001), and ML EC (P,0.001) conditions. Traditional

measures of sway speed and magnitude were not significantly lower among TC-

experts. Intention-to-treat analyses indicated no significant effects of short-term TC

training; however, increases in AP EC and ML EC complexity amongst those

randomized to TC were positively correlated with practice hours (P50.044,

P50.018). Long- and short-term TC training were positively associated with

physical function.

Conclusion: Multiscale entropy offers a complementary approach to traditional

COP measures for characterizing sway during quiet standing, and may be more

sensitive to the effects of TC in healthy adults.

Trial Registration: ClinicalTrials.gov NCT01340365

Introduction

Postural control is critical to the maintenance of balance and avoidance of falls.

Balance control systems integrate inputs from the motor cortex, cerebellum, and

basal ganglia, as well as feedback from visual, vestibular, and proprioceptive

systems to maintain upright posture. When properly functioning, this multi-level

neural control system produces stable balance and gait [1, 2]. However,

maintenance of balance, even under constant environmental conditions, involves

continuous postural adjustments that appear as irregular fluctuations in the center

of pressure (COP) displacements recorded from a balance platform [3–6]. In the

past, these fluctuations were considered to be ‘noise’ and not integral to COP

metrics. Thus, quantitative studies of balance focused on average sway or COP

parameters, ignoring temporal information. Recent research has demonstrated

that COP fluctuations convey important information. In fact, COP dynamics are

highly ‘‘complex;’’ i.e., they contain non-random fluctuations that exist across

multiple temporal and spatial scales [7, 8]. Complexity-based measures of COP

dynamics, including entropy or fractal-based metrics, may therefore be

informative outcomes for characterizing age-related decline and frailty [9], fall

risk [10] and neuromuscular disorders [1, 6, 11]. This dynamical systems

perspective of postural control is also aligned with a growing interest in evaluating

multimodal interventions to reduce fall risk, and for complexity-based metrics to

assess intervention-related changes in balance system dynamics.

health care centers). The content is solely theresponsibility of the authors and does notnecessarily represent the official views of HarvardCatalyst, Harvard University and its affiliatedacademic health care centers, or the NationalInstitutes of Health. The funders had no role instudy design, data collection and analysis, decisionto publish, or preparation of the manuscript.

Competing Interests: The authors have declaredthat no competing interests exist.

Tai Chi and Sway Complexity


http://clinicaltrials.gov/ct2/show/NCT01340365?term=NCT01340365&rank=1

Tai Chi is a multi-component mind–body exercise that is grounded in the

holistic model of traditional Chinese medicine. The explicit goals of targeting

multiple physiological, motor, and cognitive processes and integrating their

dynamics makes Tai Chi particularly well-suited as a multimodal intervention to

enhance balance within a systems biology framework [12]. Tai Chi integrates

balance, flexibility, and neuromuscular coordination training with a number of

cognitive components including relaxation, focused body awareness, imagery, and

multi-tasking, that together may result in benefits above and beyond conven-

tional, single modality exercise [13–15]. A recent Cochrane review of 159

randomized trials evaluating both exercise and non-exercise interventions

reported that Tai Chi trials had a pooled relative risk ratio of 0.71 (95% CI 0.57 to

0.87) for falls [16]. These results are supported by other systematic reviews which

conclude that Tai Chi directly reduces fall risk [17, 18] and/or positively impacts

factors associated with falling including the fear of falling [19, 20], clinical

measures of balance [21–23], musculoskeletal strength [22, 23] and flexibility

[22, 24].

The effect of Tai Chi on postural control in older adults, however, is less clear.

Whereas some studies report reductions in the average speed or magnitude of

COP fluctuations beneath the feet when standing [25, 26], others report no change

[27–30] or even increases [20, 31] in these parameters after Tai Chi training. This

is surprising, as postural instability, as evident by excessive or uncontrolled sway,

has been identified as one of the key risk factors that lead to falls in many balance-

impaired populations [32–36]. However, an increase in bodily sway could

represent a more dynamic, resilient, and adaptive postural control system; for

example in Parkinson’s disease an increase in sway may represent an increase in

the range of stability and improvement in postural control [37]. Therefore, better

measures of postural control are needed that can quantify the non-linear, dynamic

characteristics of postural stability [38].

We have proposed that the impact of Tai Chi on postural control may be better

characterized by quantifying its effects on the degree of complexity associated with

system output (i.e., COP dynamics) than by traditional sway parameters

[12, 14, 39] MSE (multiscale entropy) is one measure of physiologic complexity

that is largely independent of traditional metrics and appears to be sensitive to

aging and disease [40]. This study performs exploratory analysis of the impact of

long- and short-term Tai Chi training on postural control characterized by both

MSE and traditional COP measures in a sample of non-randomized Tai Chi

expert and randomized Tai Chi nal̈ve healthy older adults. Based on our earlier

studies, we predicted that: 1) Tai Chi experts would exhibit greater COP

complexity compared to age and gender matched Tai Chi-nal̈ve older adults; 2)

Tai Chi-nal̈ve older adults randomly assigned to 6 months of Tai Chi training

would exhibit increased COP complexity after training, as compared to a usual

care control; and 3) Complexity-based COP measures would be more sensitive to

long- and short-term exposure to Tai Chi than traditional sway measures.



Methods

Study Design

We employed a hybrid study design that included a two-arm randomized clinical

trial (RCT) along with an additional observational comparison group. The RCT

evaluated the short-term (6 months) effects of Tai Chi training on healthy

community-dwelling older adults. A group of age-matched Tai Chi experts was

compared to the randomized Tai Chi and control groups at baseline to evaluate

the potential longer-term effects of Tai Chi, and to compare the impact of short-

vs. long-term training on balance and functional outcomes. The protocol for this

trial and supporting CONSORT checklist are available as supporting information;

see S1 Checklist and S1 Protocol. The Institutional Review Board at Beth Israel

Deaconess Medical Center (BIDMC) approved this study. The study protocol was

first approved in October of 2010. The randomized trial component of this study

was registered at ClinicalTrials.gov (NCT01340365). Registration was initiated

prior to the start of recruitment, but administrative reassignment of the study staff

resulted in a 7-week delay wherein 4 participants were randomized.

Randomized trial design

A total of 60 healthy older adults, aged 50–79, were randomized to receive 6

months of Tai Chi training in addition to usual health care, or to usual health care

alone (control group). Study participants randomized to usual care were offered a

3-month course of Tai Chi as a courtesy following the trial. Randomization was

stratified by age (50–59 y, 60–69 y, 70–79 y) and utilized a permuted-blocks

randomization scheme with randomly-varying block sizes. Randomization was

performed by the study statistician. All outcomes were assessed at baseline, 3 and

6 months; primary staff overseeing assessment and analysis of balance-related and

functional assessment outcomes were blinded to treatment assignment.

Participant flow through the randomized trial component of the study is

summarized in Fig. 1. Recruitment spanned from March 2011 to March 2013. All

follow up procedures were completed by September 2013. Further details related

to the design of the RCT component of this study are reported elsewhere [12].

Recruitment for the randomized trial targeting community

dwelling healthy adults

Inclusion criteria were: 1) Age 50–79 years; 2) living or working within the Greater

Boston area; and 3) willing to adhere to 6 month Tai Chi training protocol.

Exclusion criteria were: 1) Chronic medical condition including cardiovascular

disease (myocardial infarction, angina, atrial fibrillation or presence of a

pacemaker), stroke, respiratory disease requiring daily use of an inhaler; diabetes

mellitus; active cancer (diagnosis ,5 years ago and requiring ongoing

chemotherapy or use of cytotoxic agents), stage III prostate cancer, dermatological

cancer with reoccurrence; neurological conditions (e.g., seizure disorder,

Parkinson’s, peripheral neuropathy); or significant neuromuscular or musculos-



Fig. 1. Flow of participants in randomization trial of healthy Tai Chi nal̈ve adults.

doi:10.1371/journal.pone.0114731.g001



keletal conditions requiring chronic use of pain medication; 2) acute medical

condition requiring hospitalization within the past 6 months; 3) self-reported

(current) smoking or alcohol/drug abuse; 4) uncontrolled hypertension (resting

SBP.160 or DBP.100 mm Hg); 5) abnormal heart rate (resting HR.100 bpm;

,50 bpm); 6) abnormal ECG (e.g., supraventricular tachyarrhythmia, atrial

fibrillation, significant ST wave abnormality, 2nd and 3rd degree heart block); 6)

pregnancy; 7) current use of cardio- or vaso-active drugs and medications that

can affect autonomic function including beta agonists and antagonists, drugs with

anticholinergic properties (e.g. tricyclic antidepressants or anti psychotics), and

cholinesterase inhibitors; 8) self-reported inability to walk continuously for

15 minutes unassisted; 9) regular Tai Chi practice within past 5 years; and 10)

regular participation in physical exercise on average 4 or more times per week.

Interested individuals underwent both an initial phone screen and an in-person

screen at the BIDMC Clinical Research Center. Eligible individuals provided

written informed consent and underwent baseline testing prior to randomization.

Participants within both groups were encouraged to follow usual health care as

prescribed by their primary physicians, with no limitations set by the study other

than those implicit in the eligibility criteria listed above. Participants in the Tai

Chi group received 6 months of Tai Chi training in addition to usual care. All Tai

Chi interventions were administered pragmatically at one of five pre-screened Tai

Chi schools within the Greater Boston area that met specific guidelines described

elsewhere [12, 41]. Instructors were asked to teach using the same Tai Chi style,

approach and protocols employed for non-study, community participants. Study

participants were asked to attend, on average, two classes per week over the 6

month intervention. They were also asked to practice a minimum of 30 minutes,

two additional days per week. All schools provided DVDs or printed materials to

facilitate home practice. In total, participants were asked to engage in 72 hours of

training over the 6 month intervention. Attendance at Tai Chi classes was

recorded by instructors and home practice was tracked using a weekly practice

log. Participants that reported attending a minimum of 70% of all classes and

completing 70% or more of prescribed home practice between each study visit

were considered compliant or ‘per protocol’. Adverse events were monitored via

forms provided in the participant packet requesting participants to detail the

event as well as notifying the study team within 24 hours. Additionally, all Tai Chi

schools were given adverse event forms for reporting pre-specified events that

occurred during class or at their school. Participants were called monthly by a

member of the study team to monitor adverse events as well as to encourage class

attendance and home practice.

Non-randomized comparison groups

Tai Chi experts

Twenty-seven healthy older adults (age 50–79 yrs) currently engaged in an active

Tai Chi training regimen, and each with over 5 years of Tai Chi practice were

recruited for a single observational visit. No limitation was set on Tai Chi style.



The Tai Chi expert group was used to evaluate the longer-term effects of Tai Chi

training on balance outcomes. Eligibility and screening procedures for Tai Chi

experts were identical to those for healthy adults enrolled in the RCT, with the

exceptions of no limitation on exercise activity, prior Tai Chi experience, or the

use of beta blockers to control diagnosed hypertension.

Young Comparison Group

Fifteen healthy, young Tai Chi-nal̈ve adults (age 25–35 yrs) were also recruited for

a single observational visit. The group was used to better characterize age-related

trends in balance which were required to determine parameters of complexity-

based measures of sway as described below.

Outcome measures

The pre-specified balance-related primary outcome measure for the randomized

study was 6-month change in center-of-pressure complexity. All outcome

measures were assessed at the Clinical Research Center Syncope and Falls in the

Elderly (SAFE) laboratory at Beth Israel Deaconess Medical Center (Boston, MA).

Balance related outcomes reported here were part of a larger battery of tests that

lasted an average of 3 h (including cardiovascular and gait outcomes to be

reported elsewhere).

Assessment of postural sway

Tests of quiet standing were conducted while subjects stood on a force platform

for 60 s with arms by their side, feet shoulder-width apart. Subjects were asked to

stand as still as possible and to visually fixate on an ‘‘X’’ drawn on a wall

approximately 3 m away at eye-level. COP displacement was recorded with a

standard force plate (Kistler Instruments Corp, Amherst, NY). Separate tests were

conducted with eyes open and eyes closed. To account for potential trial-to-trial

variability, two trials were completed for each condition (random sequence) with

at least one minute rest between trials. During the first trial, the position of the

hallux was marked with tape to ensure consistent foot placement throughout the

study. Both traditional and complexity-based sway parameters were estimated

from COP displacement data.

Traditional COP parameter calculations

Traditional sway parameters included anterior-posterior (AP), medio-lateral

(ML), and COP (composite AP and ML) sway velocity (mm/sec), as well as

elliptical area (mm2). Raw trajectories were extracted from the force plates and

smoothed with a 4th order Butterworth low-pass filter with cut-off frequency of

10 Hz [42]. Analyses were performed using Matlab (v7, The Mathworks, Inc.,

Natick, MA).



COP complexity calculations

COP complexity was calculated using DataDemon (Dynadx Inc., Mountain View,

CA) and Matlab. For a given COP time-series, a complexity index, CI, was

calculated from multiscale entropy (MSE) analysis. Briefly, MSE utilizes Sample

Entropy (SampEn) to quantify the degree of irregularity within a time-series

across multiple time scales [10]. To do so, a coarse-graining procedure is used,

which averages sequential data points to produce multiple new time-series, each

with a characteristic time scale. At time scale 1, no coarse-graining occurs and the

original N samples are used to calculate SampEn. This metric characterizes

irregularity by calculating the negative natural logarithm of the conditional

probability that sequences similar for m points will remain similar if one more

point is added to the sequence. Similar points have amplitudes which fall within a

percentage, r, of the time-series standard deviation of each other. In our study we

used m52 and r50.15. At each subsequent time scale factor (t) in the MSE

algorithm, averages of t points are taken within non-overlapping windows to

create a new time-series of length N/t (Fig. 2). At each time scale, t, SampEn is

calculated to produce an MSE curve (i.e., SampEn versus t). Using this output, CI

was derived by calculating the area under the curve. Since the area under the curve

is based on SampEn, a larger area or CI represents greater irregularity over a range

of time scales and thus higher complexity. As explained below, the range of scales

used to calculate CI varied based on the direction of sway (AP vs. ML) and

comparison (long-term or short-term Tai-Chi training). Therefore, we present all

complexity results as the complexity index divided by the range of MSE scales for

a particular sway direction and comparison. This represents the average CI per

MSE scale. We denote this as CI/R(t), where R(t) represents the range of MSE

scales. Normalizing the reported complexity allows comparison across both sway

directions and conditions.

Complexity pre-processing

Raw COP time-series were down-sampled to 250 Hz. Because trends at time scales

longer than those being analyzed alter estimates of CI, we detrended the COP

time-series prior to MSE analysis [10]. To do so, each time-series was detrended

using ensemble empirical mode decomposition (EEMD). EEMD is well-suited for

the decomposition of nonstationary and nonlinear signals [43]. This process

creates multiple intrinsic mode functions (IMFs), each with predominant power

over a limited frequency range. While subsequent IMFs may have spectrum

overlap, each IMF captures properties of the original time-series at different time-

scales [10]. For this analysis, thirteen IMFs were generated. We removed IMFs 9–

13 which exhibited frequencies below 0.2 Hz in order to ensure a minimum

number of dynamic patterns within the length of our time series [6]. Additionally,

we removed IMF 1 and 2 which exhibited frequencies above 20 Hz and are thus

unlikely to reflect balance-related biological processes.



Complexity mode selection

Twenty-one unique IMF combinations were generated, consisting of all

continuously sequenced combinations of IMFs 3–8. We then selected the specific

IMF combination to be used for analyses based on a statistical comparison

between groups prior to treatment, using an approach similar to Wei et al. [44].

Briefly, Wei and colleagues performed statistical tests to choose IMFs that best

discriminated young and elderly participants with known differences in balance,

and then used these IMF combinations to test the effect of stochastic resonance on

the complexity of balance in the elderly group. Similarly, to test our hypothesis

that Tai Chi experts would exhibit higher complexity than age-matched controls,

we made the a priori assumption that young adults would have higher COP

complexity than older adults, and that Tai Chi would restore age-related COP

decline. We therefore identified the IMF combination for this hypothesis by

comparing the young group and randomized older adult Tai Chi nal̈ve group at

baseline. The IMF combination that best distinguished, statistically, the young

group from the randomized older adults group was chosen. Table 1 provides a

summary of the statistical comparison between these groups at baseline for the

IMFs which best distinguished them in each sway direction (AP and ML). We

speculate that choosing the IMFs in this manner may identify the time scales of a

healthy postural control system since literature shows that young adults generally

show superior postural steadiness relative to elderly [45]. These IMFs were

subsequently selected for the long-term Tai Chi comparison between Tai Chi

experts and age-matched controls of COP complexity. To test the hypothesis that

Tai Chi nal̈ve subjects who participated in 6 months of Tai Chi training would

improve their complexity relative to age-matched controls, we made the a priori

assumption that short-term Tai Chi training would have qualitatively similar

effects as long-term training. We thus compared the Tai Chi experts and

Fig. 2. Schematic illustration of the course-graining procedure for multiscale entropy. xi represents theoriginal time-series samples. yj represents the coarse-grained samples at the indicated scale. Adapted fromCosta et al. 2002 [72].




randomized subjects at baseline, and used the IMF combinations that statistically

best distinguished the Tai Chi experts from the randomized group. Table 2

provides a summary of the statistical comparison between these groups at baseline

for the IMFs which best distinguished them. In other words, we contend that

selecting the IMF combination leading to COP complexity values that best

distinguish these two age-matched groups (Experts Vs. Nal̈ve) may be most

sensitive to a change in the complexity of postural control within the Nal̈ve group

following exposure to short-term Tai-Chi. These IMF combinations were used to

compare COP complexity of randomized subjects assigned to Tai Chi against the

controls assigned to usual care across their three visits. In both long- and short-

term comparisons, we used a two-sample t-test to produce the p-value statistic

used for this analysis. All of the comparisons used for IMF selection were done

independently for AP and ML directions under the eyes-open condition. The

statistically selected time-series of a particular IMF combination was then

analyzed using MSE as described above.

Clinical measures of balance, physical function and activity

Clinical balance and physical function outcomes were assessed to help interpret

sway measures. Single-legged standing balance was assessed according to Vereeck

et al. [46]. Three trials were completed under an eyes-closed condition, and the

greatest duration (sec) for each condition was used for analysis. This test has been

correlated with fall risk in older adults [47]. The Timed Up and Go Test (TUG)

was used to quantify functional mobility [48]. TUG has high test-retest reliability

and discriminant validity in older adults [49, 50]. Physical activity level was

assessed using Physical Activity Status Scale (PASS) [51]. Subjects were asked to

estimate their general physical activity during the previous week using an 11-point

scale (i.e., 0–10). The scale quantifies physical activity duration by a combination

of the minutes of exercise per week and the intensity of this exercise (heavy,

modest, or none). The concurrent validity of the scale has been documented in

both men and women and scores correlate with maximal oxygen consumption in

younger and older adults [52, 53].

Table 1. Details of the intrinsic mode functions (IMFs) and MSE scales used for the long-term Tai-Chi comparison between Nal̈ve and Expert groups.

COP Direction IMFsCharacteristic IMFFrequencies (Hz)

MSEScales

MSE ScaleRange Young Complexitya Naive Complexitya

p-value

AP 3+4+5+6 17.6, 7.7, 3.2, 1.3 1–25 25 1.206¡0.256 1.034¡0.237 ,0.001

ML 3+4+5+6+7 17.9, 8.0, 3.4, 1.2, 0.5 1–35 35 0.943¡0.161 0.819¡0.181 ,0.001

The last three columns show the statistical comparison between the Young and Naive at baseline which led to the selection of these IMFs.avalues provided are mean ¡ standard deviation with units of complexity index divided by MSE scale range, CI/R(t).

doi:10.1371/journal.pone.0114731.t001



Statistical methods

Baseline measures of Tai Chi masters and Tai Chi nal̈ves were compared in a

linear model controlling for age. The group difference and adjusted means for a

participant with mean age were estimated. Trends in balance measures over the

24-week intervention were compared between nal̈ve participants randomized to

Tai Chi or usual care using a random-slopes model with shared baseline. All

longitudinal analysis were conducted according to the intention-to-treat

paradigm. The model included fixed effect of time, time6treatment, age, and

time6age and random participant-specific intercepts and slopes with unstruc-

tured covariance. The shared baseline assumption, enforced by omitting a

treatment main-effect term, properly reflects the true state of the population

sampled prior to randomization and has the advantage of adjusting for any chance

differences at baseline in a manner similar to ANCOVA. A similar model with

visit treated as a categorical term was also fit to explore possible non-linearity in

the time profiles. As above, treatment-group differences and adjusted means for a

participant with mean age were estimated as well as their 95% confidence

intervals. Residual diagnostics were used to test assumptions of data normality,

and in a small number of cases where non-normality was detected, log-

transformation was employed. Log-transformation had only minor impacts on

statistical results and yielded equivalent inference, thus only non-transformed

results are presented. Associations between percent changes in traditional and

complexity-based measures of balance vs. dosage of Tai Chi training, and

functional measures vs. Tai Chi training dosage were estimated by linear

regression. All inferential tests were two-tailed at alpha50.05. We chose to report

comparison-wise p-values without adjustment for multiple comparisons to avoid

inflating type II errors, recognizing that the nominal p-values underestimate the

overall experiment-wise type I error rate. Our results are intended as hypothesis

generating, not definitive tests of efficacy for Tai Chi on any specific measure of

balance. All analyses were performed in SAS (version 9.3, SAS Institute, Cary,

NC).

Regarding sample size considerations, for cross-sectional comparisons, we

estimated that a sample size of 27 Tai Chi Expert and 60 Tai Chi nal̈ve subjects

would provide power to detect an effect size of 0.63. For the randomized trial, we

estimated that the sample of 60 participants randomized 1:1, the study would have

Table 2. Details of the intrinsic mode functions (IMFs) and MSE scales used for the short-term Tai-Chi comparison between the randomized to Tai-Chi andUsual Care groups.

COP Direction IMFsCharacteristic IMFFrequencies (Hz)

MSEScales

MSE ScaleRange

ExpertsComplexitya Naive Complexitya

p-value

AP 5+6+7 3.2, 1.3, 0.6 2–31 30 0.948¡0.136 0.843¡0.162 ,0.001

ML 4+5+6+7+8 8.0, 3.4, 1.2, 0.5, 0.3 1–39 39 0.929¡0.207 0.747¡0.180 ,0.001

The last three columns show the statistical comparison between the Experts and Naive at baseline which led to the selection of these IMFs.avalues provided are mean ¡ standard deviation with units of complexity index divided by MSE scale range, CI/R(t).




80% power to detect a main effect of treatment if the true effect size was at least

0.74 based on a two-tailed test at p,0.05.

Results

Participant characteristics

Tai Chi experts (n527) reported an average of 24.6¡12 years of Tai Chi training

experience (median 20 yrs, range 10–50 yrs). Approximately equal numbers

reported Yang (n512) and Wu (n515) style Tai Chi as their primary training

systems; however, all reported having training experience in others styles of Tai

Chi, related internal and external martial arts (e.g. kung fu, bagua) and/or mind-

body practices (e.g. yoga, meditation). Sociodemographic characteristics of the

Tai Chi experts were well-matched with the older Tai Chi nal̈ve group with respect

to average age and global cognitive status. Compared with nal̈ve older controls,

experts included a slightly greater proportion of men and Asians, and had lower

BMI’s and higher levels of physical activity (Table 3).

Tai Chi nal̈ve adults subsequently randomized to Tai Chi plus usual care vs.

usual care alone were comparable at baseline. For all variables, values for the

subset of participants that were found to be ‘per-protocol’ were comparable to

those in the larger sample thereby minimizing potential sources of bias in post-

hoc comparisons between control and Tai Chi compliant groups.

Recruitment and protocol adherence

Six hundred and seventy-nine older adults were approached in order to recruit

and enroll 60 eligible healthy adults. The majority were excluded due to medical

ineligibility (214) and limited time availability or interest (179). Of those enrolled,

97% (28/29) and 87% (27/31) of individuals in the usual care and Tai Chi group

completed the primary 6-month follow-up assessment, respectively. All 60

participants completing baseline assessments were included in intent to treat

analyses. Adherence to the Tai Chi protocol was variable. Fifteen of the 31 (48%)

participants in the Tai Chi group were protocol compliant, attending 70% of

classes and completing 70% of required home practice between each study visit.

This compliant subgroup had a mean exposure to Tai Chi training of 89.3 hours;

in comparison the intent to treat group had a mean exposure of 60.9 hours.

Overall satisfaction with Tai Chi programs, assessed on a 1–7 scale (1 is best score)

were high, with a mean ¡ standard deviation rating of 1.6¡1.1 at 3 months and

1.7¡1.4 at 6 months.

A total of 4 non-serious adverse events were reported throughout study. All

events were reported by participants randomized to the Tai Chi group. Only 2 of

the 4 events were determined to be related to the Tai Chi intervention (both

minor musculoskeletal injuries (one wrist, one ankle)).



The association between long-term Tai Chi training and standing

postural control

Complexity-based measures of both anterior-posterior (AP) and mediolateral

(ML) sway assessed under eyes open (EO) and eyes closed (EC) conditions were

consistently higher for Tai Chi experts vs. Tai Chi nal̈ve subjects (Table 4). Age

adjusted linear models revealed that average MSE of sway in the Tai Chi experts

group were markedly and statistically greater than the Tai Chi nal̈ve group under

ML EO (21%; p,0.001) and ML EC (20%; p,0.001) testing conditions, and to a

lesser degree in the AP EC (6%; p50.023) condition (see Fig. 3 a, b). A negative

overall trend with age was only observed in the AP EC condition but there were

no group6age interaction effects (Table 4).

In contrast, we found no differences between the Tai Chi nal̈ve and Tai Chi

expert group for traditional measures of sway, although 7 of 8 traditional

measures exhibited trends towards greater sway among Tai Chi experts (Table 4).

More than half of the traditional parameters increased significantly with age (age

P#0.05) and AP EO sway speed, AP EC sway speed and COP EC sway speed

exhibited a significant group6age interaction indicating a larger Tai Chi-related

Table 3. Participant characteristics for observational and randomized groups.

Observational Groups Randomized Groups and Sub-groups

Older Tai ChiExperts

Older Tai ChiNal̈ve

Young Tai ChiNal̈ve Usual Care Tai Chi TC Compliant

(n527) (n560) (n515) (n529) (n531) (n515)

Agea 62.78¡7.57 64.18¡7.68 28.73¡3.20 64.45¡7.42 63.94¡8.02 63.47¡6.98

Gender n(%)

Male 13 (48.1%) 20 (33.3%) 8 (53.3%) 11 (37.9%) 9 (29%) 5 (33.3%)

Female 14 (51.9%) 40 (66.7%) 7 (46.7%) 18 (62.1%) 22 (71%) 10 (66.7%)

Race n(%)

White 22 (81.5%) 55 (91.7%) 15 (100%) 26 (89.7%) 29 (93.5%) 15 (100%)

African American 1 (3.7%) 3 (5%) 0 (0%) 3 (10.3%) 0 (0%) 0 (0%)

Asian 4 (14.8%) 2 (3.3%) 0 (0%) 0 (0%) 2 (6.5%) 0 (0%)

Ethnicity n(%)

Non-Hispanic/Non-Latino 26 (96.3%) 59 (98.3%) 14 (93.3%) 29 (100%) 30 (96.8%) 14 (93.3%)

Hispanic/Latino 1 (3.7%) 1 (1.7%) 1 (6.7%) 0 (0%) 1 (3.2%) 1 (6.7%)

Education (yrs)a 18.44¡3.34 16.7¡3.25 17.93¡2.52 16.19¡3.03 17.13¡3.41 17.43¡2.85

MMSE (out of 30)a 29.07¡1.11 29.12¡1.01 n/a 29.21¡0.82 29.03¡1.17 28.93¡1.28

BMI (kg/m2)a 23.54¡2.35 26.46¡5.46 25.43¡3.21 26.54¡5.83 26.38¡5.19 26.81¡4.60

Physical Activity Levela,b 6.0¡2.0 4.4¡2.2 n/a 4.0¡2.0 4.0¡2.0 4.9¡2.2

Single leg stance time (eyes closed) (sec)a 19.6¡14.07 8.56¡6.27 24.82¡15.23 7.52¡5.74 9.53¡6.67 8.57¡5.79

Timed Up and Go (sec)a 5.07¡0.86 5.94¡1.14 4.43¡0.68 5.68¡0.99 6.17¡1.24 6.11¡1.11

avalues provided are mean ¡ standard deviation.bPhysical Activity Level descriptions:45Run about 1 mile per week OR walk about 1.3 miles per week OR spend about 30 minutes per week in comparable physical activity.55Run about 1 to 5 miles per week OR walk 1.3 to 6 miles per week OR spend 30 to 60 minutes per week in comparable physical activity.65Run about 6 to 10 miles per week OR walk 7 to 13 miles per week OR spend 1 to 3 hours per week in comparable physical activity.




difference among older participants. Adjusting for baseline BMI and exercise

activity in linear models did not substantially alter the results of the complexity

measures for ML EO or EC, but did modestly reduce significance levels for AP EC

complexity (p50.147, p50.086 for BMI and exercise activity, respectively).

Education level, MMSE, and TUG did not impact any model results.

The effect of short-term Tai Chi training on standing postural

control

Intention-to-treat analyses using age adjusted linear models revealed no

statistically significant group6time interactions for any complexity-based

measures of sway, although complexity assessed under the EC condition trended

towards higher values in the Tai Chi group (Table 5). Analyses limited to per-

protocol Tai Chi subjects indicated a significantly greater increase in complexity

in the ML EC condition in the Tai Chi group compared to usual care (P50.029).

Additionally, associations between changes in complexity score and Tai Chi

dosages (combined total hours of class and home training) were all positive and

statistically significant under AP EC and ML EC conditions (Fig. 4a–d). Of note,

sway complexity levels following 6 months of Tai Chi training did not reach the

complexity levels of Tai Chi experts under all testing conditions (P,0.03 for all

measurements).

Age adjusted linear models indicated no statistically significant group6time

interactions for 7 of the 8 traditional sway parameters, but as with cross-sectional

Table 4. Complexity-based and Traditional measures of sway for Nal̈ve and Expert groups at baseline.

Outcome Groups Statistical Result

Nal̈ve (n560) Expert (n527)

Complexity COP Measures Mean ¡ S.D.a Mean ¡ S.D.a p - ageb p – groupc

AP - Eyes Open (CI/R(t)) 1.033¡0.216 1.107¡0.209 0.510 0.131

AP - Eyes Closed (CI/R(t)) 0.982¡0.211 1.082¡0.200 0.028 0.023

ML - Eyes Open (CI/R(t)) 0.819¡0.161 0.993¡0.217 0.353 ,0.001

ML - Eyes Closed (CI/R(t)) 0.834¡0.188 1.001¡0.219 0.574 ,0.001

Traditional COP Measures

AP Sway Speed - Eyes Open (mm/s) 6.59¡1.90 7.09¡2.26 0.040 0.203

AP Sway Speed - Eyes Closed (mm/s) 9.94¡3.61 10.83¡3.99 0.050 0.227

ML Sway Speed - Eyes Open (mm/s) 3.71¡1.65 3.84¡2.19 0.143 0.656

ML Sway Speed - Eyes Closed (mm/s) 4.72¡2.46 4.95¡4.12 0.050 0.624

COP Sway Speed - Eyes Open (mm/s) 8.29¡2.45 8.85¡3.21 0.037 0.276

COP Sway Speed - Eyes Closed (mm/s) 11.97¡4.29 12.92¡6.31 0.028 0.296

Elliptical Area - Eyes Open (mm2) 161.8¡136.5 170.4¡332 0.083 0.748

Elliptical Area - Eyes Closed (mm2) 200.3¡135.6 166.4¡126.8 0.148 0.330

avalues provided are mean ¡ standard deviation.bp-values for testing for an effect of age.cp-values for comparing Nal̈ve vs. Expert groups after adjusting for age.




comparisons of Tai Chi experts vs. Tai Chi nal̈ves, values for all 8 parameters

showed trends towards increasing sway with exposure to Tai Chi (Table 5). These

trends were not affected by limiting analyses to compliant per-protocol Tai Chi

subjects. Adding physical activity into longitudinal models for both traditional

and complexity based COP measures did not significantly affect results. Of note,

in contrast to complexity measures, values of most traditional sway parameters

following 6 months of Tai Chi were very similar to values of Tai Chi experts.

Impact of short- and long-term training on physical function

Both single leg stance time (SLST) and Timed Up and Go (TUG) performance

were better in the Tai Chi expert group as compared to the Tai Chi nal̈ve group

(P,0.001; Fig. 5). Changes over 6 months in functional performance showed

trends towards being greater in those randomized to Tai Chi vs. usual care, but

group6time interactions were not significant in either intent-to-treat or per-

Fig. 3. Complexity-based and traditional measures of sway for age-matched Tai Chi nal̈ve and Tai Chi expert older adults. Mean value and standarderrors for sway measured in the anterioposterior (AP) and mediolateral (ML) planes and during eyes-open (EO) and eyes-closed (EC) conditions. Statisticalcomparisons were adjusted for age.




Table

5.Complexity-base

dandTraditionalmeasu

resofsw

ayfortheRandomizedto

Tai-ChiandRandomizedto

usu

alca

regroupsacross

visits.

COPOutcome

Measures

Groups

Statistical

Result

Tai-Chi

UsualCare

Tai-Chi

Usual

Care

Baseline

3Months

6Months

Baseline

3Months

6Months

AcrossVisit

Across

Visit

Complexity

Mean

¡S.D.a

Mean

¡S.D.a

Mean

¡S.D.a

Mean

¡S.D.a

Mean

¡S.D.a

Mean

¡S.D.a

Slope

(95%

CI)b

Slope

(95%

CI)

bp-

agec

p–

grp*td

AP-EO

(CI/R

(t))

0.831¡0.124

0.843¡0.123

0.824¡

0.148

0.857¡0.167

0.845¡

0.171

0.811

¡0.163

20.008

(20.031,

0.015)

20.017

(20.039,

0.006)

0.951

0.499

AP-EC

(CI/R

(t))

0.835¡0.156

0.865¡0.176

0.857¡

0.145

0.878¡0.129

0.863¡

0.138

0.846¡0.127

0.007

(20.011

,0.025)

20.010

(20.028,

0.008)

0.070

0.117

ML-EO

(CI/R

(t))

0.706¡0.167

0.731¡0.202

0.686¡

0.171

0.792¡0.162

0.757¡

0.161

0.742¡0.159

20.019

(20.046,

0.007)

20.012

(20.038,

0.014

0.580

0.605

ML-EC

(CI/R

(t))

0.743¡0.177

0.806¡0.215

0.787¡

0.175

0.838¡0.160

0.826¡

0.158

0.777¡0.138

0.007

(20.019,

0.033)

20.013

(20.038,

0.013)

0.914

0.166

Traditional

APSS-EO

(mm/s)

6.33¡

1.72

6.86¡2.35

6.75¡1.98

6.86¡

2.07

6.76¡2.07

6.72¡2.16

0.119

(20.124,

0.361)

20.006

(20.247,

0.235)

0.497

0.432

APSS-EC

(mm/s)

10.00¡3.97

10.57¡3.76

10.77¡

4.53

9.88¡

3.25

9.98¡3.13

9.88¡3.13

0.295

(20.075,

0.666)

0.033

(20.334,

0.400)

0.744

0.289

MLSS-EO

(mm/s)

3.73¡

1.54

3.81¡1.44

3.96¡1.88

3.68¡

1.79

3.33¡1.45

3.55¡1.46

0.154

(20.032,

0.340)

20.054

(20.238,

0.129)

0.479

0.077



Table

5.Cont.

COP

Outcome

Measures

Groups

Statistical

Result

Tai-Chi

UsualCare

Tai-Chi

Usual

Care

Baseline

3Months

6Months

Baseline

3Months

6Months

AcrossVisit

Across

Visit

Complexity

Mean

¡S.D.a

Mean

¡S.D.a

Mean

¡S.D.a

Mean

¡S.D.a

Mean

¡S.D.a

Mean

¡S.D.a

Slope

(95%

CI)b

Slope

(95%

CI)

bp-

agec

p–

grp*td

MLSS-EC

(mm/s)

4.92¡2.55

4.61¡1.72

4.97¡

2.66

4.50¡2.40

4.05¡

1.76

4.11¡1.59

0.105

(20.147,

0.356)

20.202

(20.451,

0.046)

0.317

0.034

COP

SS-

EO

(mm/s)

8.10¡2.15

8.60¡2.85

8.63¡

2.79

8.49¡2.75

8.19¡

2.60

8.31¡2.51

0.223

(20.084,

0.530)

20.027

(20.330,

0.277)

0.432

0.213

COP

SS-

EC

(mm/s)

12.15¡4.76

12.44¡

4.07

12.88¡5.40

11.78¡3.80

11.56¡3.63

11.53¡3.35

0.352

(20.088,

0.791)

20.095

(20.530,

0.341)

0.537

0.119

Ellip.Area-

EO

(mm

2)

172.7

¡125.1

217.6

¡211

.1250.5

¡270.7

150.2

¡149.1

137.6

¡147.4

141.1

¡103.9

35.9

(5.09,

66.8)

1.88

(228.7,

32.5)

0.596

0.051

Ellip.Area-

EC

(mm

2)

234.3

¡140.5

213.7

¡177.8

287.4

¡275.7

163.9

¡122.2

154.1

¡92.6

173.8

¡100.2

33.3

(3.20,63.4)

1.07

(229.1,

31.2)

0.226

0.078

ava

luesprovidedare

mean

¡standard

deviatio

n.

bva

luesare

estim

atedslope(lower95%

confid

ence

interval,upper95%

confid

ence

interval).

cp-valuesfortestingforaneffe

ctofageindicatedbyp-age.

dp-valuesforco

mparingmeanratesofch

angeamongTa

i-Chivs

.usu

alca

regroupsindicatedbyp-group*tim

e.




protocol analyses. However, associations between changes in functional scores vs.

Tai Chi dosage (combined total hours of class and home training) were positive

and statistically significant for both SLST and TUG (Fig. 6). Of note, levels of

physical function following 6 months of Tai Chi training were lower than that

seen in the Tai Chi experts (p50.064 and p50.001 for SLST and TUG,

respectively).

Discussion

Maintaining balance and preventing falls among older people is an urgent public

health challenge worldwide with significant direct and indirect costs to society

[54–56]. This challenge has led to extensive efforts to improve our understanding

of the complex physiology underlying age-related decline of postural control, to

Fig. 4. Associations between changes in both complexity-based (A, B) and traditional measures (C, D) of sway vs. Tai Chi dosage (total hourspracticed) among Tai Chi nal̈ve individuals exposed to six months of training. Percent changes in sway were assessed between the baseline and 6month follow up visit for the anterioposterior (AP) and mediolateral (ML) planes and during eyes-open (EO) and eyes-closed (EC) conditions. Analysesexcluding the largest x-axis value (i.e. potential outlier) reduced DAP EC p-value to a non-significant level (p50.423), however, the p-value for the DML ECwas less affected (p50.037).




identify interventions to reduce the risk of falling in older adults, and to develop

markers to accurately measure the short- and long-term impact of interventions

targeting balance and fall risk. Mounting evidence suggests that Tai Chi is a

promising intervention for reducing falls, especially in older (transitionally) frail

Fig. 5. Measures of physical function for age-matched Tai Chi nal̈ve and Tai Chi expert older adults.Mean value and standard errors for A) Single leg stance time (SLST) and B) timed up and go (TUG).Comparisons were adjusted for age.




and neurologically impaired adults that are at high risk of falling [16, 37, 57, 58].

However, less is known about Tai Chi’s impact on healthy adults that are at a

lower risk for falling, the relevance of measures of sway for evaluating

Fig. 6. Associations between changes in physical function and Tai Chi dosage (total hours practiced)among Tai Chi nal̈ve individuals exposed to six months of training. Percent changes in function wereassessed between the baseline and 6 month follow-up visit for single leg stance time (SLST) and timed upand go (TUG) parameters. Analyses excluding the largest x-axis value (i.e. potential outlier) reduced DSLSTto a non-significant level (p50.286), however, the p-value for the DTUG was less affected (p50.0062).




intervention-related impacts on postural control, and the mechanisms through

which Tai Chi may reduce future fall risks.

To our knowledge, this study represents the first randomized trial evaluating

the impact of Tai Chi on complexity-based measures of postural sway. Our main

finding is that complexity measures may provide a complementary, alternative

and perhaps more discriminating metric for characterizing postural sway during

quiet standing and the impact of complex mind-body interventions such as Tai

Chi on balance. Among healthy adults, long-term exposure to Tai Chi training

was generally associated with increased sway complexity. Intent-to-treat analyses

did not indicate a significant effect of shorter-term (6 month) exposure to Tai Chi

training on sway complexity or function. However, post-hoc analyses indicated

positive and significant associations between hours of exposure to Tai Chi training

and increases in sway complexity and functional performance. In combination

with prior research linking enhanced complexity with health (discussed below),

these findings suggest that MSE may be a sensitive maker for characterizing the

benefits of Tai Chi to postural control during quiet standing in healthy adults. In

contrast, traditional measures of sway did not reflect obvious effects of either

long- or short-term Tai Chi training. In fact, although trends were not statistically

significant, the majority of traditional sway outcomes increased with exposure to

Tai Chi. Increased sway in many, but not all (see below), populations has been

associated with increased fall risk [32–36]. Our findings suggest that these

traditional sway metrics may be relatively insensitive to the effects of Tai Chi in

healthy active older adults, and may also require alternative interpretations.

The findings in the current study parallel those of a pilot observational study

evaluating the effects of 24 weeks of Tai Chi for individuals with peripheral

neuropathy. Tai Chi training significantly increased MSE-derived complexity of

standing COP dynamics, and complexity indices were associated with improved

plantar sensation and physical function [39]. In contrast, no intervention-related

changes in traditional sway parameters were observed.

Although higher values of traditional sway metrics have been identified as one

of the key risk factors that lead to falls [32–36], the effects of Tai Chi on

traditional sway metrics is variable and unclear. Some studies report reductions in

the average speed or magnitude of COP fluctuations [25, 26, 59], others report no

change [27–30], and similar to the trends we observed in this study, some studies

have reported increases [20, 31] in traditional sway parameters after Tai Chi

training. Of note, the effects of conventional therapies on COP metrics have been

variable, sometimes increasing following interventions [60]. Some of this

variability may result from the use of testing conditions that vary from quiet

standing to more provocative challenges to balance. It is also possible that this

reported variability in the effects of Tai Chi on traditional sway parameters, and

perhaps also the insensitivity in response to Tai Chi we observed in the present

study, reflects a diversity of strategies for postural control and associated

reductions in fall risks. A hallmark study by Wolf and colleagues [61] compared

the effect of 15 weeks of computerized balance training to Tai Chi on standing

postural sway and fall risk in older frail adults. Balance training, but not Tai Chi,



reduced the average magnitude and velocity of sway. Under certain conditions,

average sway magnitude was actually greater following Tai Chi training.

Interestingly, however, the risk of falling was significantly reduced only in the Tai

Chi group (adjusted RR 0.51 vs. 0.98 for Tai Chi and balance training,

respectively). Wolf and colleagues [20] suggested that Tai Chi-related changes in

fall risk may be due to factors other than average sway characteristics. They also

suggested that the inherent aim of Tai Chi training might not be to reduce sway,

but rather to provide corrective skills and improve confidence for managing

postural instability. Other studies also support that Tai Chi may increase the

limits of stability within which individuals are comfortable [31, 37, 62, 63]. Finally,

it is possible that trends towards higher sway may be related to reduced

musculoskeletal rigidity, as evidence supports that Tai Chi training reduces

muscle co-contraction, and increases flexibility [64, 65]. This could produce

beneficial effects in conditions such as Parkinson’s disease [37]. This perspective is

indirectly supported by studies of individuals with ligament laxity and

hypermobility (Ehlers-Danlos syndrome) who exhibit markedly greater COP sway

ranges when compared to healthy controls [66]. The same study also reported that

complexity based metrics of COP (approximate and sample entropy) were

markedly lower in the hypermobile group, interpreted as a lower level of

integrated postural control. Together, these results suggest that the non-linear

measures we explored here, which focus on quantifying the moment-to-moment

control of postural sway during relatively unchallenged quiet standing conditions,

may afford unique and complementary insight into Tai Chi’s effect on

mechanisms of postural control that are not well-captured by traditional

measures.

There is growing evidence to suggest that greater complexity of sway is

associated with higher levels of function and health. One recent study [9]

evaluated the COP dynamics from 550 participants (avg. 77.9 yrs) whose frailty

phenotype (not frail, pre-frail, or frail) was determined using standard criteria.

Complexity of COP dynamics was quantified using MSE. Compared to the non-

frail, MSE of COP was lower in pre-frail and frail groups (p,0.002). Although

traditional linear measures of balance (root mean square amplitude of sway) were

also associated with frailty, statistical models indicated that only MSE

independently predicted frailty status after accounting for other physiologic

determinants of balance such as age, vision, lower extremity strength, peripheral

neuropathy, and frontal-executive cognitive function. MSE apparently quantifies

an aspect of frailty not captured by other measures.

Complexity-based analyses of COP have been shown to be sensitive indicators

of pathology. Cavanaugh et al. [67] used approximate entropy (ApEn) to examine

changes in postural stability in 27 collegiate athletes following cerebral

concussions (all underwent baseline testing prior to their athletic season and their

injury). COP dynamics were also measured in a cohort of age-matched healthy

non-athletes. None of the athletes with concussions exhibited any indications of

postural instability using standard testing. However, ApEn values, a measure

similar to SampEn, following concussions generally declined, whereas ApEn



values remained stable for healthy subjects. Complexity markers are now being

considered as part of a post-concussion assessment protocol [5].

A number of studies have shown that the complexity of COP time series

declines with age [4, 10, 68, 69]. For example, Costa et al. [10] evaluated short-

term COP dynamics in 15 healthy young adults (avg. 27 y), 22 healthy elderly

adults (avg. 75 y), and 22 elderly fallers (avg. 74 y). For sway velocity measures,

complexity indices in both the AP and ML plane were significantly different for all

three groups. In the current study, we did not observe any significant age or

treatment6age effects on complexity. This may reflect a lack of association

between MSE measures of sway and aging, or it may reflect a bias in our study

sampling design. Because our eligibility criteria for Tai Chi nal̈ve older adults was

quite narrow with respect to acceptable comorbidities and use of medications, our

sample reflects a very healthy adult population. As the number of individuals that

met our eligibility criteria decreased with age, those in the oldest stratified age

group (70–79 yrs) were more likely to represent extremely healthy individuals,

whereas those on our younger group (50–59 yrs) that met our eligibility criteria

were more likely to be of average health. This interpretation is supported by our

TUG data. Average TUG scores for our 60–70 and 70–80 y group were 6.1 and

6.2 seconds; in comparison, normative data for ambulatory adults in these two

age groups is 8 and 9 seconds. Due to this bias, our ability to evaluate the direct

effects of age on COP complexity, as well as how age may impact response to Tai

Chi is partially limited.

Strengths and Limitations

There are a number of strengths and limitations to this study. One strength is our

hybrid design and our ability to compare long- and short-term training of Tai

Chi. Multiple prior cross-sectional studies have compared older Tai Chi experts

vs. older Tai Chi nal̈ve adults with respect to traditional measures of balance and

function [70, 71]. However, to our knowledge, our study is the first to compare

the effects of short-term Tai Chi training to long-term training, using both

traditional and novel measures of postural control, and employing a prospective

randomized trial design with identical protocols to our cross-sectional study. Our

hybrid design also has important limitations. Samples for both our cross-sectional

comparisons and our RCT were small, and could have resulted in type II errors.

Additionally, as with any cross-sectional study, comparisons between Tai Chi

experts and Tai Chi nal̈ve may be confounded by differences between groups other

than exposure to Tai Chi. While linear models that included potential

confounders (e.g., BMI, exercise activity) suggested an independent effect of long-

term Tai Chi even after these factors were taken into account, other factors,

including training in other mind-body exercises (qigong) and martial arts could

not be fully accounted for. For these reasons, caution should be used when

attributing any cross-sectional differences solely to the effect of long-term Tai Chi

training. The lack of an active exercise control in our RCT also leaves open the

possibility that the effects we observed were primarily due to psychosocial



interactions. Future studies comparing Tai Chi vs other exercise controls on

complexity measures of sway will be informative. Future comparative studies

should also consider combining detailed measures of physiological processes

known to impact postural control with complexity based measures. This approach

would facilitate the understanding of relative contributions of specific

physiological processes and their relationship to specific IMFs, and thus the

pathways through which interventions like Tai Chi impact balance. Finally, in this

exploratory study, we did not make statistical adjustments for multiple

comparisons, which increases the likelihood that we observed some positive

treatment related effects simply due to chance.

Conclusions

This study represents the first evaluation of the utility of complexity-based

measures of postural sway for characterizing the impact of short- and long-term

mind-body exercise training, namely Tai Chi. Multiscale entropy offers a

complementary approach to traditional COP measures for characterizing sway

during quiet standing, and may be more sensitive to the effects of Tai Chi in

healthy adults. Future appropriately powered studies should explore how

complexity vs. traditional measures of sway correlate with validated measures of

functional performance and help predict critical public health outcomes including

long-term risk of injurious falls.

Supporting Information

S1 Checklist. CONSORT Checklist.

doi:10.1371/journal.pone.0114731.s001 (DOC)

S1 Protocol. Trial Protocol.

doi:10.1371/journal.pone.0114731.s002 (DOCX)

Acknowledgments

We thank Danielle Berkowitz and Mary Quilty for research assistance.

Author ContributionsConceived and designed the experiments: PMW BM JMH CKP RBD LAL GYY

ACA. Performed the experiments: JNW ML BM. Analyzed the data: BJG MDC

PMW. Contributed to the writing of the manuscript: PMW BJG BM JNW LAL

JMH MDC CKP RBD ML VN EAM GYY ACA. Participated in clinical screening

of study subjects and evaluated EKG’s: GYY ACA.



http://www.plosone.org/article/fetchSingleRepresentation.action?uri=info:doi/10.1371/journal.pone.0114731.s001

http://www.plosone.org/article/fetchSingleRepresentation.action?uri=info:doi/10.1371/journal.pone.0114731.s002

References

1. Hausdorff JM (2007) Gait dynamics, fractals and falls: finding meaning in the stride-to-stride fluctuationsof human walking. Hum Mov Sci 26: 555–589.

2. Shumway-Cook A, Woollacott M (2006) Motor Control: Translating Research into Clinical Practice:Philadelphia, Lippincott Williams and Wilkins. 576 p.

3. Collins JJ, De Luca CJ (1993) Open-loop and closed-loop control of posture: a random-walk analysis ofcenter-of-pressure trajectories. Exp Brain Res 95: 308–318.

4. Duarte M, Sternad D (2008) Complexity of human postural control in young and older adults duringprolonged standing. Exp Brain Res 191: 265–276.

5. Cavanaugh JT, Guskiewicz KM, Stergiou N (2005) A nonlinear dynamic approach for evaluatingpostural control: new directions for the management of sport-related cerebral concussion. Sports Med35: 935–950.

6. Manor B, Costa MD, Hu K, Newton E, Starobinets O, et al. (2010) Physiological complexity andsystem adaptability: evidence from postural control dynamics of older adults. J Appl Physiol 109: 1786–1791.

7. Costa M, Goldberger AL, Peng CK (2002) Multiscale entropy analysis of complex physiologic timeseries. Phys Rev Lett 89: 068102.

8. Peng CK, Costa M, Goldberger AL (2009) Adaptive data analysis of complex fluctuations in phsyiologictime series. Adv Adapt Data Anal 1: 61–70.

9. Kang HG, Costa MD, Priplata AA, Starobinets OV, Goldberger AL, et al. (2009) Frailty and thedegradation of complex balance dynamics during a dual-task protocol. J Gerontol A Biol Sci Med Sci 64:1304–1311.

10. Costa M, Priplata AA, Lipsitz LA, Wu Z, Huang NE, et al. (2007) Noise and poise: Enhancement ofpostural complexity in the elderly with a stochastic-resonance-based therapy. Europhys Lett 77: 68008.

11. Hausdorff JM (2009) Gait dynamics in Parkinson’s disease: common and distinct behavior among stridelength, gait variability, and fractal-like scaling. Chaos 19: 026113.

12. Wayne PM, Manor B, Novak V, Costa MD, Hausdorff JM, et al. (2013) A systems biology approach tostudying Tai Chi, physiological complexity and healthy aging: design and rationale of a pragmaticrandomized controlled trial. Contemp Clin Trials 34: 21–34.

13. Burgener SC, Yang Y, Gilbert R, Marsh-Yant S (2008) The effects of a multimodal intervention onoutcomes of persons with early-stage dementia. Am J Alzheimers Dis Other Demen 23: 382–394.

14. Wayne P, Kaptchuk T (2008) Challenges inherent to Tai Chi research: Part I–Tai Chi as a complex multi-component intervention. J Altern Complement Med 14: 95–102.

15. Wayne P, Kaptchuk T (2008) Challenges inherent to Tai Chi Research: Part II–Defining the interventionand optimal study design. J Altern Complement Med 14: 191–197.

16. Gillespie LD, Robertson MC, Gillespie WJ, Sherrington C, Gates S, et al. (2012) Interventions forpreventing falls in older people living in the community. Cochrane Database Syst Rev 9: CD007146.

17. Harmer PA, Li F (2008) Tai Chi and falls prevention in older people. Med Sport Sci 52: 124–134.

18. Liu H, Frank A (2010) Tai chi as a balance improvement exercise for older adults: a systematic review.Journal of Geriatric Physical Therapy 33: 103–109.

19. Sattin RW, Easley KA, Wolf SL, Chen Y, Kutner MH (2005) Reduction in fear of falling through intensetai chi exercise training in older, transitionally frail adults. J Am Geriatr Soc 53: 1168–1178.

20. Wolf SL, Barnhart HX, Ellison GL, Coogler CE (1997) The effect of Tai Chi Quan and computerizedbalance training on postural stability in older subjects. Atlanta FICSIT Group. Frailty and Injuries:Cooperative Studies on Intervention Techniques. Phys Ther 77: 371–381.

21. Manor B, Lough M, Gagnon MM, Cupples A, Wayne PM, et al. (2014) Functional benefits of Tai Chitraining within senior housing facilities. Journal American Geriatric Society: In Press.

22. Jacobson BH, Chen HC, Cashel C, Guerrero L (1997) The effect of T’ai Chi Chuan training onbalance, kinesthetic sense, and strength. Percept Mot Skills 84: 27–33.



23. Judge JO, Lindsey C, Underwood M, Winsemius D (1993) Balance improvements in older women:effects of exercise training. Phys Ther 73: 254–262.

24. Lan C, Lai JS, Chen SY, Wong MK (1998) 12-month Tai Chi training in the elderly: its effect on healthfitness. Med Sci Sports Exerc 30: 345–351.

25. Guan H, Koceja DM (2011) Effects of Long-Term Tai Chi Practice on Balance and H-ReflexCharacteristics. American Journal of Chinese Medicine 39: 251–260.

26. Wu G, Zhao F, Zhou X, Wei L (2002) Improvement of isokinetic knee extensor strength and reduction ofpostural sway in the elderly from long-term Tai Chi exercise. Arch Phys Med Rehabil 83: 1364–1369.

27. Chen Y-S, Zhou S, Cartwright C (2011) Effect of 12 weeks of Tai Chi training on soleus Hoffmann reflexand control of static posture in older adults. Arch Phys Med Rehabil 92: 886–891.

28. Li L, Manor B (2010) Long term Tai Chi exercise improves physical performance among people withperipheral neuropathy. American Journal of Chinese Medicine 38: 1–11.

29. Tsang WW, Hui-Chan CW (2003) Effects of tai chi on joint proprioception and stability limits in elderlysubjects. Med Sci Sports Exerc 35: 1962–1971.

30. Tsang WW, Wong VS, Fu SN, Hui-Chan CW (2004) Tai Chi improves standing balance control underreduced or conflicting sensory conditions. Arch Phys Med Rehabil 85: 129–137.

31. Pluchino A, Lee SY, Asfour S, Roos BA, Signorile JF (2012) Pilot study comparing changes inpostural control after training using a video game balance board program and 2 standard activity-basedbalance intervention programs. Arch Phys Med Rehabil 93: 1138–1146.

32. Kenney RA, Rubenstein LZ, Martin FC, Tinetti ME (2001) Guideline for the prevention of falls in olderpersons. American Geriatrics Society, British Geriatrics Society, and American Academy of OrthopaedicSurgeons Panel on Falls Prevention. J Am Geriatr Soc 49: 664–672.

33. Maki BE, Holliday PJ, Topper AK (1994) A prospective study of postural balance and risk of falling in anambulatory and independent elderly population. J Gerontol 49: M72–84.

34. Melzer I, Benjuya N, Kaplanski J (2004) Postural stability in the elderly: a comparison between fallersand non-fallers. Age Ageing 33: 602–607.

35. Pajala S, Era P, Koskenvuo M, Kaprio J, Tormakangas T, et al. (2008) Force platform balancemeasures as predictors of indoor and outdoor falls in community-dwelling women aged 63–76 years.J Gerontol A Biol Sci Med Sci 63: 171–178.

36. Rubenstein LZ, Powers CM, MacLean CH (2001) Quality indicators for the management andprevention of falls and mobility problems in vulnerable elders. Ann Intern Med 135: 686–693.

37. Li F, Harmer P, Fitzgerald K, Eckstrom E, Stock R, et al. (2012) Tai chi and postural stability in patientswith Parkinson’s disease. N Engl J Med 366: 511–519.

38. Lipsitz LA (2002) Dynamics of stability: the physiologic basis of functional health and frailty.J Gerontol A Biol Sci Med Sci 57: B115–125.

39. Manor B, Lipsitz LA, Wayne PM, Peng CK, Li L (2013) Complexity-based measures inform tai chi’simpact on standing postural control in older adults with peripheral neuropathy. BMC Complement AlternMed 13: 87.

40. Costa M, Goldberger AL, Peng CK (2005) Multiscale entropy analysis of biological signals. PhysRev E Stat Nonlin Soft Matter Phys 71: 021906.

41. Fischer M, Fugate-Woods N, Wayne PM (2014) Use of pragmatic community-based interventions toenhance recruitment and adherence in a randomized trial of Tai Chi for women with osteopenia: insightsfrom a qualitative substudy. The North American Menopause Society 21: In Press.

42. Robertson DG, Dowling JJ (2003) Design and responses of Butterworth and critically damped digitalfilters. J Electromyogr Kinesiol 13: 569–573.

43. Wu Z, Huang NE (2009) Ensemble Empirical Mode Decomposition: A Noise-Assisted Data AnalysisMethod. Advances in Adaptive Data Analysis 1: 1–41.

44. Wei Q, Liu D, Wang K, Liu Q, Abbod MF, et al. (2012) Multivariate Multiscale Entropy Applied to Centerof Pressure Signals Analysis: An Effect of Vibration Stimulation of Shoes. Entropy 14: 2157–2172.



45. Prieto TE, Myklebust JB, Hoffmann RG, Lovett EG, Myklebust BM (1996) Measures of posturalsteadiness: differences between healthy young and elderly adults. IEEE Trans Biomed Eng 43: 956–966.

46. Vereeck L, Wuyts F, Truijen S, Van de Heyning P (2008) Clinical assessment of balance: normativedata, and gender and age effects. Int J Audiol 47: 67–75.

47. Kuh D, Bassey EJ, Butterworth S, Hardy R, Wadsworth ME (2005) Grip strength, postural control,and functional leg power in a representative cohort of British men and women: associations with physicalactivity, health status, and socioeconomic conditions. J Gerontol A Biol Sci Med Sci 60: 224–231.

48. Podsiadlo D, Richardson S (1991) The timed ‘‘Up & Go’’: a test of basic functional mobility for frailelderly persons. J Am Geriatr Soc 39: 142–148.

49. Lin MR, Hwang HF, Hu MH, Wu HD, Wang YW, et al. (2004) Psychometric comparisons of the timed upand go, one-leg stand, functional reach, and Tinetti balance measures in community-dwelling olderpeople. J Am Geriatr Soc 52: 1343–1348.

50. Shumway-Cook A, Brauer S, Woollacott M (2000) Predicting the probability for falls in community-dwelling older adults using the Timed Up & Go Test. Phys Ther 80: 896–903.

51. Baumgartner T, Jackson A, Mahar M, Rowe D (2006) Measurement for Evaluation in PhysicalEducation and Exercise Science: New York, McGraw-Hill 560 p.

52. Heil DP, Freedson PS, Ahlquist LE, Price J, Rippe JM (1995) Nonexercise regression models toestimate peak oxygen consumption. Med Sci Sports Exerc 27: 599–606.

53. Jackson AS, Beard EF, Wier LT, Ross RM, Stuteville JE, et al. (1995) Changes in aerobic power ofmen, ages 25–70 yr. Med Sci Sports Exerc 27: 113–120.

54. Rubenstein LZ, Josephson KR (2002) The epidemiology of falls and syncope. Clin Geriatr Med 18:141–158.

55. Schneider EC, Shubert TE, Harmon KJ (2010) Addressing the escalating public health issue of fallsamong older adults. N C Med J 71: 547–552.

56. Stevens JA, Corso PS, Finkelstein EA, Miller TR (2006) The costs of fatal and non-fatal falls amongolder adults. Inj Prev 12: 290–295.

57. Tousignant M, Corriveau H, Roy PM, Desrosiers J, Dubuc N, et al. (2012) Efficacy of supervised TaiChi exercises versus conventional physical therapy exercises in fall prevention for frail older adults: arandomized controlled trial. Disabil Rehabil 35: 1429–1435.

58. Taylor-Piliae RE, Hoke TM, Hepworth JT, Latt LD, Najafi B, et al. (2014) Effect of Tai Chi on physicalfunction, fall rates and quality of life among older stroke survivors. Arch Phys Med Rehabil 95: 816–824.

59. Wu G, Keyes L, Callas P, Ren X, Bookchin B (2010) Comparison of telecommunication, community,and home-based Tai Chi exercise programs on compliance and effectiveness in elders at risk for falls.Arch Phys Med Rehabil 91: 849–856.

60. Lima BN, Lucareli PR, Gomes WA, Silva JJ, Bley AS, et al. (2014) The acute effects of unilateralankle plantar flexors static- stretching on postural sway and gastrocnemius muscle activity during single-leg balance tasks. J Sports Sci Med 13: 564–570.

61. Wolf SL, Barnhart HX, Kutner NG, McNeely E, Coogler C, et al. (1996) Reducing frailty and falls inolder persons: an investigation of Tai Chi and computerized balance training. Atlanta FICSIT Group.Frailty and Injuries: Cooperative Studies of Intervention Techniques. J Am Geriatr Soc 44: 489–497.

62. Gyllensten AL, Hui-Chan CWY, Tsang WWN (2010) Stability limits, single-leg jump, and bodyawareness in older Tai Chi practitioners. Arch Phys Med Rehabil 91: 215–220.

63. Tsang WW, Hui-Chan CW (2004) Effect of 4- and 8-wk intensive Tai Chi Training on balance control inthe elderly. Med Sci Sports Exerc 36: 648–657.

64. Gatts S (2008) Neural mechanisms underlying balance control in Tai Chi. Med Sport Sci 52: 87–103.

65. Manson J, Rotondi M, Jamnik V, Ardern C, Tamim H (2013) Effect of tai chi on musculoskeletalhealth-related fitness and self-reported physical health changes in low income, multiple ethnicity mid toolder adults. BMC Geriatr 13: 114.



66. Rigoldi C, Cimolin V, Camerota F, Celletti C, Albertini G, et al. (2013) Measuring regularity of humanpostural sway using approximate entropy and sample entropy in patients with Ehlers-Danlos syndromehypermobility type. Res Dev Disabil 34: 840–846.

67. Cavanaugh JT, Guskiewicz KM, Giuliani C, Marshall S, Mercer V, et al. (2005) Detecting alteredpostural control after cerebral concussion in athletes with normal postural stability. Br J Sports Med 39:805–811.

68. Collins JJ, De Luca CJ, Burrows A, Lipsitz LA (1995) Age-related changes in open-loop and closed-loop postural control mechanisms. Exp Brain Res 104: 480–492.

69. Thurner S, Mittermaier C, Ehrenberger K (2002) Change of complexity patterns in human postureduring aging. Audiol Neurootol 7: 240–248.

70. Hong Y, Li JX, Robinson PD (2000) Balance control, flexibility, and cardiorespiratory fitness amongolder Tai Chi practitioners. Br J Sports Med 34: 29–34.

71. Hakim RM, Kotroba E, Cours J, Teel S, Leininger PM (2010) A cross-sectional study of balance-related measures with older adults who participated in Tai Chi, yoga, or no exercise. Physical &Occupational Therapy in Geriatrics 28: 63–74.

72. Costa M, Goldberger AL, Peng CK (2002) Multiscale entropy to distinguish physiologic and syntheticRR time series. Comput Cardiol 29: 137–140.



Complexity-Based Measures Inform Effects of Tai Chi Training on Standing Postural Control:...

Documents

Transcript of Complexity-Based Measures Inform Effects of Tai Chi Training on Standing Postural Control:...